Introduction

Drought is a major abiotic stress that restricts crop growth, development and yield and thus has turned into a grave threat to universal food security [1]. In addition, global climate change, particularly high temperatures and erratic rainfall patterns, combined with a growing world population, is placing tremendous stress on food security and sustainability. These challenging conditions can be overcome through breeding programs to develop drought-resistant crops [2, 3]. The combined effect of drought and other abiotic stresses can reduce potential crop production by more than 50%. According to modeling simulations, drought-affected crop** regions could quadruple by the end of the twenty-first century [4]. In response to these conditions, rice employs various adaptive methods, such as building up various osmoprotectants or solutes and changes in the direction of plant growth to avoid drought [5]. Roots are the plant’s main organs that anchor it in the soil and are required for nutrition and water absorption. Favorable responses of plants to water stress is dependent on the roots’ capacity to maintain growth (i.e., modifying the root traits such as depth, density, and root angles) and maintain/increase root hydraulic conductivity [6,7,8]. The ability of roots to tolerate water deficiency depends on their ability to maintain adequate carbohydrate metabolism, cell wall protein composition, osmotic potential, and metabolites involved in the oxidative stress response [9, 10]. Most research has been focused on improving features in above-ground tissues to tolerate these pressures, but roots (the ‘hidden half’ of a plant’s architecture) remain an underutilized source of crop development [11]. Root System Architecture (RSA) is critical for improving nutrient and water uptake and maintaining crop yield under optimal and drought conditions [12]. Extensive root systems can help plants overcome drought, which is influenced by growth angle, root thickness, and length [13, 2). The score plot of the first two PCs is shown in Fig. 1A. The majority of the data fell inside the 92% confidence interval (Hotelling T2 ellipse). The PCA results for the four cluster samples showed that there was a clear distinction between the control (c) and treated samples (drought-stressed), while no clear difference was seen between the two genotypes (Fig. 1A). To confirm this trend, three of the whole competitive groups were examined by PCA, yielding similar results. The PCA models yielded two, two, and three PCs, respectively, for comparing AZs vs. AZc, IRs vs. IRc, and AZs vs. IRs samples, respectively. The R2 X and Q2 (goodness of prediction) values are shown in Table 2 and indicate that the differences between groups could be predictably explained by all the models. However, no distinct limit could be seen between the two groups’ PCA score plots (Table 2). PLS-DA is a supervised method that categorizes the observations into groups that yield the largest predicted indicator variable. The obtained data resulted in two PCs (R2 X = 0.586, R2 Y = 0.512, Q2 = 0.471) between the four cluster, and enhanced the classification between these groups in the score plot (Fig. 1B). It led to better modeling and prediction results with two PCs, when the data were examined with only the control and the treated-AZ or treated-IR64 (R2 Y > 0.9, Q2 > 0.7) (Table 2) As shown in Table 2, the treated AZ and IR samples could be separated with two PCs in spite of the slight overlap in the PLS-DA score plot (R2 Y > 0.5, Q2 > 0.08), indicating intrinsic metabolic differences between these two genotypes in the treated conditions.

Table 2 Explanation and predictability values of the principal component analysis (PCA) and partial least squares-discriminate analysis (PLS-DA)
Full size table
Fig. 1
figure 1

Principal component analysis (PCA) and Partial least squares-discriminate analysis (PLS-DA) score plots of metabolic profiles in rice roots under drought stress. A PCA score plot for Azucena normal (red), IR64 normal (blue), Azucena-treated (yellow) and IR64-treated (green) samples, B PLS-DA score plot for Azucena normal (red), IR64 normal (blue), Azucena-treated (yellow) and IR64-treated (green) samples

Full size image

Overview of Azucena and IR64 root-tips metabolome under control and drought stress conditions

The line plots of the X-loadings of the first component of the PLS-DA pairwise comparison models were used to identify the primary altered metabolites. Based on the parameter VIP > 1, p-value ≤ 0.01, log fold change (log FC > 2.0), and Kruskal–Wallis ANOVA, a total of 103 drought responsive metabolites with significant differences were identified (Table S2 and S3). The most significant metabolites had variable importance in the projection (VIP) values greater than 1, which was reported to explain the responses [30]. The VIP values for metabolites categorized by superclass are shown in Fig. 2A. Amino acid and organic acid groups were most abundant. For example, GABA and aspartic acid had a VIP value of 20.52 and 18.30, although the average value of the exclusive VIP was 5.63. Organic acid (16%), amino acid (12.2%), polyphenols (12.8%), nucleic acid derivatives (8.3%) and unclassified (others, 7.1%) were the most common differentially abundant metabolites (Fig. 2B).

Fig. 2
figure 2

Classification of the differentially abundant metabolites and their variable importance in the projection (VIP) distribution in IR64 and Azucena in root tips. A VIP distribution in each metabolite superclass as a scatter plot. The average mean of the differentially abundant metabolites is shown in the red dashed line (B) A pie chart depicting the proportion of each metabolite in the superclass

Full size image

Compared to the control condition, the levels of 56 metabolites increased and 24 metabolites decreased in the Azucena root-tips in response to drought stress. In contrast, in IR64, the levels of 30 metabolites increased, and 19 decreased in response to drought stress (Fig. 3A). A cross-comparison of the differentially abundant metabolites between genotypes showed that number of accumulated metabolites were nearly two times higher in the root tips of the tolerant genotype (53 metabolites) than the sensitive one (22 metabolites), and of which 27 metabolites were commonly altered between the two genotypes in response to drought stress (Fig. 3B). The 25 most differentially abundant metabolites are shown in Fig. 4. The log2 fold change values and VIP score for these metabolites are mentioned in Table S2 & Table S3. As shown in the Fig. 4, the levels of amino acids including aspartic acid and glutamic acid, nucleotides namely thymine and guanine increased in both genotypes under drought stress. Interestingly, glycine, phenylalanine, threonine, isoleucine, and GABA were accumulated in IR64, while there was no significant difference in the levels of these metabolites in Azucena under drought stress. Likewise, there is a considerable increase in the levels of TCA cycle intermediates comprising of malic acid, isocitric acid, succinic acid and fumaric acid, and sugar and sugar alcohol such as sorbitol, mannitol, galactinol, myo-inositol, D-raffinose, sucrose, ribose and trehalose. In particular, the levels of several metabolites involved in secondary metabolism comprising of genistein, vanillin, scopoletin, conifery aldehyde, farnesyl pyrophosphate, betaine, cyanidin 3-O-rutinoside 5-O-beta-D-glucoside, and syringic acid showed the highest levels in the drought tolerant cultivar Azucena.

Fig. 3
figure 3

Overview of differentially abundant metabolites (DAMs) between the Azucena and IR64. A Up- and down-regulated metabolites of Azucena and IR64 in response to drought stress. B Venn diagram showing the overlap between DAMs responsive to drought stress

Full size image
Fig. 4
figure 4

A Bar graph of 25 differentially abundant metabolites in Azucena root tips. VIP values are in a blue column and the red columns represent log2 (fold change, FC) values. B Bar graph of 25 differentially abundant metabolites in IR64 root tips. VIP values are in a blue column and the red columns represent log2 (fold change, FC) values

Full size image

Identification of potential association between metabolites and observed root traits

Metabolite content was determined in the root tips of control and drought-stressed plants from two different genotypes. The hierarchical clustering for both metabolites and samples (genotypes × conditions) is shown in Fig. 5. Clustering of the samples showed complete separation of the metabolite pattern between the control and drought-treated samples. Thus, drought treatment was the main source of variance in the data, indicating a complete change in metabolism under stress conditions in both genotypes. Figure 5 shows the metabolites that changed significantly under drought stress; glutamic acid, aspartic acid, proline, glucose-6-phosphate, and thymine were among the predominant metabolites that increased in response to drought stress, whereas metabolites belonging to quinic acid and ribonic acid (lowest group in Fig. 5) decreased under drought stress.

Fig. 5
figure 5

Metabolite response to drought differs between the two rice cultivars. Hierarchical clustering and heatmap of metabolite levels in root tip of IRc: IR64 genotype under control condition, IRs: IR64 genotype under drought stress conditions, AZc: Azucena genotype under control conditions, AZs: Azucena genotype under drought stress conditions

Full size image

The potential correlation between the abundant metabolites and the root phenotype were tested by analyzing the correlations of expression of the metabolite levels with phenotypic traits. We rely on the prediction marker for high concentrations since high amounts of a metabolite can be detected more accurately than low concentrations or their absence. High metabolite concentrations in tolerant cultivars are indicated by significant positive correlations of metabolites with phenotypic traits, whereas high concentrations in susceptible cultivars are indicated by significant negative correlations. In a positive correlation, the metabolite would be a tolerance marker because its higher concentrations would contribute to the tolerance. Metabolites with negative correlation are sensitivity markers. Most of the significant correlations between metabolite levels and root phenotypic parameters were found to be positive under drought stress (Fig. 6). Positive correlations were observed for the concentration of the trehalose, proline, succinic acid, tryptophan, salicylic acid, sucrose, fructose lysine and mannitol. Higher concentrations of these metabolites were associated with the number of tips, root length (cm), surface area (cm2), and root diameter. In tolerant plants as opposed to sensitive plants, concentrations of these metabolites were higher during drought stress.

Fig. 6
figure 6

Correlation of root phenotypic data with metabolite levels. Correlation coefficients for selected metabolites with significant (p≤0.05) positive (blue) or negative (red) correlation between concentration metabolite levels with root phenotypic data under drought or control conditions. Data of root tip of two cultivars (Azucena and IR64) grown in two conditions. Mean values of three to five replicates per cultivar and condition were correlated. ARA (Analyzed Region Area (cm2)); ARW (Analyzed Region Width (cm)); ARH (Analyzed Region Height (cm)); RL (Root Length (cm)); SA (Surface Area (cm2)); N.Tip ( number of tips), LR ( lateral root (%)), RD (root diameter (cm)

Full size image

However, for most of these metabolites, no correlations were found between concentrations under control conditions and root performance under drought. In contrast, concentrations of ribose under control conditions correlated positively with performance under drought and since there was a positive correlation between concentrations and root traits performance under control conditions also, ribose concentrations appear to be related to rate of root growth rather than drought tolerance. Gallic acid and ascorbic acid made better candidates for drought markers because their concentrations exhibited positive correlation with root performance only under drought stress conditions. Negative correlations were found for concentrations of leucine, isoleucine, pyroglutamic acid, phenylalanine, and glycerol. Higher concentrations of these metabolites were associated with decreased root length, surface area, and number of root tips. However, GABA concentration was strongly correlated with root diameter. The levels were 10 to 100 times higher in the sensitive genotype than in the tolerant genotype.

Functional annotation of DAMs

The differentially altered metabolites were functionally categorized based on the DAMs in Azucena and IR64, respectively, according to the KEGG database (www.kegg.jp/kegg/kegg1.html). When the control and treated plants were compared, it was observed that drought stress significantly altered the relative abundance of the levels of several metabolite classes. The most represented categories were organic acid compounds, amino acids, polyphenols, flavonoids, and sugars. Between the Azucena and IR64 genotypes, there was a significant variation in the proportion of organic acid compounds and biosynthesis of carbohydrates and amino acids. The metabolites in Azucena were mainly associated with amino acid biosynthesis and the TCA cycle, aminoacyl-tRNA biosynthesis, and fatty acid biosynthesis (Fig. 7) while metabolites in the IR64 genotype were mainly related to phenylalanine and galactose metabolism (p < 0.05) (Fig. 8). Increase in sucrose, glucose, tryptophan, and proline in the metabolic pathways of Azucena, may have a substantial impact on how resistant Azucena is to drought. Drought stress can have severe consequences for phenylalanine production in IR64. The high GABA expression together with its negative string correlation with root traits also indicates that IR64 restricts root elongation. Candidate DAMs demonstrating important functions or variant-specific expression profiles in Azucena and IR64 are listed in Table S2 and S3, and their relationships to major functional categories are shown in Fig. 9. A total of 103 DAMs were enriched in 30 metabolic pathways, which were most strongly represented by starch and sucrose metabolism, amino acid biosynthesis, secondary metabolite biosynthesis, purine metabolism, and fatty acid biosynthesis.

Fig. 7
figure 7

KEGG pathway classification of the differentially abundant metabolites under drought stress in Azucena (www.kegg.jp/kegg/kegg1.html)

Full size image
Fig. 8
figure 8

KEGG pathway classification of the differentially abundant metabolites under drought stress in IR64 (www.kegg.jp/kegg/kegg1.html)

Full size image
Fig. 9
figure 9

Primary metabolism responses in two rice genotypes (Azucena and IR64) after drought stress. Colors depict the relative accumulation levels of each metabolite: white (no significance), red (increase) and blue (decrease)

Full size image

Expression of genes involved in drought stress response

Potential marker genes involved in sugar metabolism, root growth and elongation were identified in our previous transcriptome studies on two rice genotypes with different drought tolerance [31, 32]. Thirteen genes were selected as candidate markers, and their expression levels were measured by qRT-PCR in the roots of the two genotypes grown under control and drought stress conditions. It was found that the gene Ethylene response factor (ERF35) was significantly expressed in the Azucena genotype (5.63-fold), whereas it was less expressed in the IR64 genotype (3.56-fold). Gene expression of serine/threonine kinase (SnRK2), IAA19, Trehalose -phosphate phosphate1 (OsTPP1), SUCROSE TRANSPORTER 5 (SUT5), dehydration responsive element binding protein (DREB) increased (upregulated) in Azucena by 4.86-fold, 5.38-fold, 5.90-fold, 8.45-fold and 6.75-fold, respectively, while they had no detectable change in IR64 (Fig. 10). Primer sequences and more details about genes can be found in Table S6. To find the interrelation between expression levels of selected genes and metabolites, the correlation analysis was done. The connections between metabolite levels and expression levels were in the positive rather than in the negative direction indicating a co-regulation at both the transcriptional and metabolite levels. For instance, there was a significant positive correlation between the level of proline and expression of the pyrroline-5-carboxylate reductase (P5CR) which in involved in proline metabolism (Fig. 11). Further positive connections were found between PRP5, glycine- and proline-rich protein 3 (OsGPRP3), trehalose phosphate phosphate1 (TPP1), sucrose transport protein (SUT5), and auxin-responsive protein IAA19 (IAA19) with trehalose, sucrose, and tryptophan, which were present at high levels under drought condition (Fig. 11). Thus, the high expression of these genes indicates drought tolerance under drought stress conditions.

Fig. 10
figure 10

Validations of selected genes using qRT-PCR in root tip zone of both genotypes, Azucena and IR64, in response to water stress

Full size image
Fig. 11
figure 11

Correlation of the expression level of selected genes with the abundance of metabolites. Correlation coefficients for selected metabolites with significant (p ≤ 0.05) positive (blue) or negative (red) correlation between metabolite concentration and expression level of selected gene data under drought conditions

Full size image

Discussion

Contrasting drought tolerances of the two rice genotypes

The effectiveness of water uptake from a diverse soil environment is determined by root architecture. In drought condition due to increased soil resistance and decreased water availability, roots are unable to absorb or release water to the soil when the soil dries out. This leads to a decrease in osmotic potential and matric potential [33], as well as a decrease in turgor pressure and cell volume [34]. Root cells need to develop techniques to counteract water loss and its consequences. At times, the solution potential of the cells is reduced, increasing turgor pressure and allowing development to continue under water-deficient conditions [104].

Quantitative gene expression analysis

Consistent with our previous miRNA-seq and mRNA reports [31, 32], 13 genes involved in root growth and elongation and regulation of root primary metabolism were selected. Total RNA isolation was done using the TRIzol reagent (BioBasic-BS410A, Canada) as per the manufacturer’s guidelines. RNA quality was assessed on an agarose gel (1%). RNA quantity was found using a NanoPhotometer® spectrophotometer (NP80 NanoPhotometer, IMPLEN). The cDNA was reverse transcribed from the isolated RNA using the SuperScript First-Strand System for the RT-qPCR, which was performed using Invitrogen™ Kit. PerlPrimer v.1.1.21 software was used for primer designing from the transcribed region of the rice genes (sequences obtained from the RAP-DB database). qRT-PCR was accomplished using SYBR Green Master Mix (Eurogentec, Köln, Germany) in the ABI Prism 7900HT (Applied Biosystems, Foster City, CA) with the usual thermal cycling conditions (95 °C for 10 min, 95 °C for 15 s for 40 cycles, 60 °C for 1 min). The experiment was conducted in three biological replicates and two technical replicates were used. For the purpose of examining the dissociation curves for shoulders or extra peaks, the SDS 2.2.1 software (Applied Biosystems) was utilized. The expression values were normalized using the UBQ (Ubiquitin) gene as a housekee** gene [115]. LinRegPCR was used to calculate primer efficiency [116]. “Normalized expression of the genes of interest was calculated by dividing the average relative expression (primer efficiency P to the power of cycle number Ct) of the housekee** genes (H1) by the relative expression of the gene of interest (GOI): (GOI): ((PH1^CtH1)/2)/PGOI^CtGOI.”